Enhanced fuzzy partitions vs data randomness in FCM
Enhanced fuzzy partitions vs data randomness in FCM
复制标题
FCM 中增强的模糊分区与数据随机性
DOI：
10.3233/ifs-141130
复制
发表时间：
2014-01-01
影响因子：
2
通讯作者：
Wang, Shitong
中科院分区：
文献类型：
--
作者：
Jiang, Yizhang;Chung, Fu-Lai;Wang, Shitong
IFP-FIM和GIFP-FCM是两种典型的增强型模糊聚类算法，其中模糊聚类的基本原理及其对噪声和/或离群值的鲁棒性通过使属于一个聚类的每个数据点的最大模糊隶属度尽可能大而属于所有其他聚类的该点的其他模糊隶属度尽可能小来增强。在本研究中，将揭示一个新的发现，即它们的增强模糊分区可以通过使用随机噪声人为地干扰给定数据集，然后将所提出的抗噪声模糊聚类算法NR-FCM应用于人为添加随机噪声的数据集来等效地实现。NR-FCM被设计为我们观察这一发现的中间步骤。这一发现的优点在于，它确实帮助我们从另一个角度见证了模糊聚类和数据随机性中模糊分区的模糊性可以协同甚至相互转化，而不是竞争。我们的几个实验结果验证了上述说法。
IFP-FIM and GIFP-FCM are two typical enhanced fuzzy clustering algorithms in which the rationale of fuzzy clustering and its robustness to noise and/or outliers are enhanced by making the maximal fuzzy membership of each data point belonging to a cluster become as big as possible and other fuzzy memberships of this point belonging to all other clusters become as small as possible. In this study, a new finding will be revealed that their enhanced fuzzy partitions can be equivalently achieved by factitiously disturbing the given dataset using a random noise and then applying the proposed noise-resistant fuzzy clustering algorithm NR-FCM to the dataset with factitiously added random noise. NR-FCM is designed as an intermediate step for us to observe this finding. The virtue of this finding exists in that it indeed helps us witness from an alternative perspective that fuzziness of fuzzy partitions in fuzzy clustering and data randomness can be collaborative and even mutually transformable rather than competitive. Our several experimental results verify the above claim.